[ScalaDays 2019] Preserving Privacy with Scala
At ScalaDays Lausanne 2019, Manohar Jonnalagedda and Jakob Odersky, researchers turned industry innovators, unveiled a Scala-powered approach to secure multi-party computation (MPC) at EPFL’s 10th anniversary conference. Their talk, moments before lunch, captivated attendees with a protocol to average salaries without revealing individual data, sparking curiosity about privacy-preserving applications. Manohar and Jakob, from Inpher, detailed a compiler transforming high-level code into secure, distributed computations, addressing real-world challenges like GDPR-compliant banking and satellite collision detection, earning applause and probing questions on security and scalability.
A Privacy-Preserving Protocol
Manohar opened with a relatable scenario: wanting to compare salaries with Jakob without disclosing personal figures. Their solution, a privacy-preserving protocol, lets three parties—Manohar, Jakob, and their CTO Dmitar—compute an average securely. Each generates three random numbers summing to zero, sharing them such that each party holds a unique view of partial sums. In Scala, this is modeled with a SecretValue type for private integers and a SharedNumber list, accessible only by the corresponding party. Each sums their shares, publishes the result, and the final sum reveals the average without exposing individual salaries. This protocol, using random shares, ensures no single party can deduce another’s data unless all communications are intercepted, balancing simplicity and security.
Secure Multi-Party Computation
Jakob explained MPC as a cryptographic subfield enabling joint function computation without revealing private inputs. The salary example used addition, but MPC supports multiplication via Beaver triplets, precomputed by a trusted dealer for efficiency. With addition and multiplication, MPC handles polynomials, enabling linear and logistic regression or exponential approximations via Taylor polynomials. Manohar highlighted Scala’s role in modeling these operations, with functions for element-wise addition and revealing sums. The protocol achieves information-theoretic security for integers, where masked values are indistinguishable, but floating-point numbers require computational security due to distribution challenges. This flexibility makes MPC suitable for complex computations, from machine learning to statistical analysis, all while preserving privacy.
Real-World Applications
MPC shines in domains where data sensitivity or legal constraints, like GDPR, restrict sharing. Manohar cited ING, a bank building credit-scoring models across European countries without moving user data across borders, complying with GDPR. Another compelling case involved satellite operators—American, Russian, or Chinese—secretly computing collision risks to avoid incidents like the 2009 Iridium-Cosmos crash, which threatened the International Space Station. Jakob emphasized that Inpher’s XOR platform, legally vetted by Baker McKenzie, ensures GDPR compliance by keeping data at its source. These use cases, from finance to defense, underscore MPC’s value in enabling secure collaboration, with Scala providing a robust, type-safe foundation for protocol implementation.
Building a Compiler for MPC
To scale MPC beyond simple embeddings, Manohar and Jakob’s team developed a compiler at Inpher, targeting a high-level language resembling Python or Scala for data scientists. This compiler transforms linear algebra-style code into low-level primitives for distributed execution across parties’ virtual machines, verified to prevent data leaks. It performs static analysis to optimize memory and communication, inferring masking parameters to minimize computational overhead. For example, multiplying masked floating-point numbers risks format explosion, so the compiler uses fixed-point representations and statistical bounds to maintain efficiency. The output, resembling assembly for MPC engines, manages memory allocation and propagates data dimensions. While currently MPC-focused, the compiler’s design could integrate other privacy techniques, offering a versatile platform for secure computation.
Links:
Hashtags: #ScalaDays2019 #Scala #Privacy #MPC